AI can make choosing easier without making markets more open. The test is whether buyers gain dependable alternatives—and whether smaller sellers can afford to serve them.
The promise of an AI shopping assistant is a market that comes to you: describe what you need, receive sensible choices, stop wasting an evening comparing tabs. An October report from PYMNTS Intelligence adds a complication. In its August survey of 2,191 US consumers, roughly 60% of respondents reporting their latest AI-assisted purchase completed it on Amazon. This is self-reported behaviour, not proof that AI increased Amazon’s sales or power. But it makes a consequential distinction visible: changing the adviser need not change who gets the order.
We should judge AI shopping by the dependable choices it creates and the livelihoods it can sustain. A fluent recommendation is only the beginning of a transaction.
Imagine a parent buying school shoes. An assistant might compare sizes, explain durability and keep the bill within a budget. Those are worthwhile gains, especially when time is scarce. Yet a recommendation that ignores delivery charges, an unreliable size chart or an expensive return can turn saved minutes into lost money. A household with little spare cash cannot treat a delayed refund as a minor inconvenience.
People also distinguish help from surrendering control. An EY Australia survey published in September, conducted in June with the Australian Food and Grocery Council, found that 72% of 1,000 adults were comfortable with AI gathering grocery prices and options; only 2% preferred an agent to complete the whole shop autonomously. These are stated Australian preferences, not measured Indian adoption. They suggest that useful assistance can stop well short of taking over.
Discovery is only half the market
For a small seller, being recommended could open a door previously guarded by advertising budgets. But someone must maintain accurate stock, dispatch the parcel, answer the complaint and fund the return. AI can reduce the effort of writing a catalogue without making these obligations disappear. Retail work may shift towards keeping promises made by an automated interface. That is a plausible change in tasks, not evidence of vanished jobs.
The economics depend on how many businesses charge for bringing buyer and seller together. If an assistant takes a referral fee on top of marketplace, payment and delivery charges, a merchant could gain orders while losing room to pay staff or improve service. If better matching reduces wasted advertising and returns, both merchant and customer could gain. The relevant measure is the margin left after the order is successfully served.
There is a strong defence of familiar marketplaces: people choose them because reliability matters. Saved payment details, predictable delivery and usable returns are valuable services. Forcing buyers towards unfamiliar merchants would mistake smaller size for better service.
The question is whether alternatives get a fair chance to demonstrate equal value. India’s Competition Commission study of AI identifies rankings that favour a platform’s own businesses as a potential route to restricted choice. That is a warning about a mechanism, not a finding against any particular shopping assistant. A recommendation funded by sellers can still help buyers, but its commercial incentives should be visible.
India has a concrete place to test this distinction. ONDC’s buyer-application requirements cover searching across the network, ranking results, displaying catalogues, combining sellers in a cart and completing payment. Buyer-facing support is part of the role. These are design requirements, not proof that small shops already prosper.
An AI interface over such a network could help a local merchant reach customers without owning the whole shopping journey. Yet an open transaction network could still acquire a narrow entrance if one assistant controls whose offers people see. Shared infrastructure does not by itself guarantee diverse recommendations.
Three futures follow. In a better one, assistants compare delivered prices and service honestly, and smaller merchants compete through reliable fulfilment. In an uneven one, consumers enjoy easier research while established checkout relationships retain most orders. In a harmful one, opaque recommendations and accumulating fees make sellers dependent on another intermediary. None is inevitable; each follows from choices about ranking, payment and responsibility.
Questions for founders and investors
Would the recommendation change if the seller paid no referral fee? Can customers compare the full delivered price, reject substitutions and obtain a refund without navigating several companies? Does a small merchant earn a viable margin after every fee and return?
Founders should test those questions on completed orders. Investors should ask for repeat purchases, complaint resolution and merchant earnings alongside growth. Network operators should make entry and ranking rules understandable, while buyers retain clear spending limits. The social promise of AI shopping rests on a simple outcome: more people can make a good purchase, and more capable sellers can make a living supplying it.
Sources